TECHNOLOGY REPORT Published jointly by Gilder Publishing, LLC and Forbes Magazine www.gildertech.com November 2001/Vol. VI No.11

The Finance of the Fibersphere

This was the month that Telecosm debtors faced their bankers and bond- holders, and Taliban warriors defected into the arms of virgins in paradise. It was the month that your alert uniformed government economists descried at last the looming monster of global deflation that has been prowling the world for nearly half a decade. For this feat of detection, the economists used pow- Well financed erful new state-of-the-art Keynesian and monetarist technologies now being power com- prepared for modular retrofit in airport luggage scanners. Although the new pany Dynegy recognition that falling prices can wreak major destruction does not in itself address the problem, the redoubtable supply-sider David Malpass believes that has contrived liquidity is on its way and the monetary drought averted, along with depression. a global In the face of it all, we perambulated a Telecosm conference around the coun- network over try. By all reports it burst through as our most cosmic and telic ever. Beset at first with Delphic doubts and airport queues, the inverted virtual towers of Nasdaq and the last year the pits of Battery Park, the itinerant confab straggled forth initially on September that in some 11 at the Inn at Squaw Creek near Tahoe with Carver Mead, Nick Tredennick and Dr. Judy Canfield nearly alone reporting for duty. respects Declaring a quorum, we mustered our tiny towering platoon at Bullwacker’s already rivals Lounge, where Carver and Nick deployed at least three quarters of our usual Global Telecosm firepower. There, crouched below the glower of a television screen unreeling every ten minutes the fiery murderous loop of images from the murky air Crossing’s of Lower Manhattan, Carver noted: “At the outset, demonic sects have it all over democratic nations. What terrorists have going for them is chiefly our ignorance, our lack of knowledge of who they are and where and what they are doing. That is an information problem, addressable by information technology. And that’s what we have, that’s what we are, and that is why we will win.” Promising a speech on the subject, he instructed us to reconvene in early November in San Francisco to devel- op an agenda for the new era. In This Issue: Over many years we have found it pays to do what Carver says. So sure enough, on November 4, while Governor Gray Davis warned of Cover Story: Mead discourses on terror; bombs on the bridges, we climbed Nob Hill with our decimated army Corvis amps up Broadwing; PCS sprints and invaded and occupied the Fairmont at the top. During a crys- ahead; Deflation destruction; Damodaran’s talline autumn week, with all bridges glowing peacefully in the sun, the dark side; Risky business; GX’s real business conference ignited a stream of flares from the summit of the Telecosm. begins; Dynegy challenges GX Two weeks later an echo of meteor showers answered from the heav- Wasteland Page 4 ens and Global Crossing (GX) shares vaulted above a dollar fifty. The only destruction came from deflation—the 40 percent appreci- Telecosm Table Page 8 ation of money over five years against all commodities—which inflict- ed devastation on all endebted companies. Already beset next generation hearing aids, Impinj, a Telecosm star by rivals plunging over technology cliffs of cost, debtors had from Washington state that is putting adaptive analog to pay back loans with funds way more valuable than those onto a Moore’s law slope, and a stealthy startup in they borrowed, while cash flow shriveled with contracting speech recognition. global business and with prices declining in all industry In comments as sage and trenchant as ever, Mead precincts unprotected by a solicitous Congress and the opened the conference with a discourse on the technolo- Pentagon. Sucked out swiftly was all capital destined for the gies needed to find and filter terror. Crucial would be fertile and competitive frontiers of the Telecosm. new applications of neural networks that could descry patterns sparsely spread through huge troves of data. With two years of rate hikes and Already used for credit card security and currency trad- warnings against irrational exuber- ing, these devices could offer real time scrutiny of every- one boarding a plane or entering a building. Combined ance, Greenspan temporarily turned with fingerprint or iris scanners, neuromorphic devices off the lights of the Telecosm could enable ready authentication of identity for many contingencies. Mead also proffered a portrait of every Once paragons of long term investment, even venture guest, swiftly captured by his Foveon camera lurking in a capitalists deployed cash mattresses, and went to sleep room off the lobby. With the pellucid pictures from a on them, dreaming of their overflowing money market CMOS silicon image plane ultimately scalable with depository accounts and keeping Milton Friedman awake Moore’s law (think of disposable cameras some day with with worries of the inflationary portent of expanding liq- ten times the resolution of high end Kodachrome) came uidity. Since debt is always crucial to ambitious infra- an implicit stock market cue. Owner of some 49 percent structure projects, deflation targeted the prime endeav- of Foveon is National Semiconductor. our of the decade: the levered delivery of the Internet For Telecosm companies, the counterpart of virgins in across vast reaches of the electromagnetic spectrum, paradise is exponential Internet traffic growth. from infrared streams down a hundred million miles of Contributing as much as anything to the stock market optical fiber, through the microwaves of a new generation mania over bandwidth companies was the estimate of of cable, cellular, and last mile wireless, and on to the John Sidgmore of WorldCom/UUNet in 1997 that traffic storewidth spans where optics and electronics merge and was doubling every 100 days, or roughly ten times a year. mesh in information mines, global caches, warehouses, New data confirms Sidgmore’s claim for those vertigi- server farms, hubs, nodes, and data centers. nous two years, when Netscape browsers, corporate With two years of interest rate hikes, inflation alarms, email, and flat rate pricing ignited the rocket. Implying and warnings against irrational exuberance, Alan a thousandfold expansion in three years and a millionfold Greenspan had temporarily turned off the lights of the in six, the Sidgmore trend was obviously unsustainable, Telecosm. Down with Loral (LOR) CDMA satellites ($3 and it was not sustained. But even a threefold annual billion in debt) and truncate Metromedia Fiber (MFNX) rise, as suggested by recent reports, would boost Internet ($4 billion). Fibrillate the nets of Broadwing (BRW) traffic 244 fold over the next five years and more than ($4.5 billion). Depress WorldCom (WCOM) as a dire 2000 fold over the next seven years. This kind of bonan- monopoly threat ($34 billion in liabilities). RIP Exodus za means huge demand for new optical and other net- (EXDS)($8 billion debt) and Globalstar (GSTRF)($3 bil- working equipment of all kinds. lion). Let’er rip Global Crossing ($6 billion). But surf- ing high on cash, with paltry debt after a six year boom, Corvis amps up Broadwing are most of the semiconductor companies on the list, led The big news from Telecosm came from Larry Roberts, by Texas Instruments (TI) with $3 billion in cash, the key founder of the ARPANET, now technical chief of an Analog Devices(ADI) with $2.5 billion, Altera (ALTR) emergent terabit router company called Caspian Networks. with $1 billion, and Atmel (ATML) with $750 million. Roberts confirmed the potential bonanza, finding that traf- Even partial chip house Terayon (TERN) commands fic had indeed risen at a rate near 2.8 times yearly until 1999 some $500 million to go with its full DOCSIS certifica- and 2000 when it ratcheted to a annual pace of nearly four- tion for next generation cable modems. National fold, which it maintained through April of this year. Semiconductor (NSM) with $822 million of liquid Although traffic growth did flatten briefly in May, perhaps as assets may find its perfect wave in Foveon. a result of the suppression of some 9.6 petabytes per month of Napster traffic (about one fourth of the Internet total), a Mead discourses on terror new surge erupted between June and October, with a bulge Foveon brings us back to its chairman and founder, after September 11. Carver Mead. The Caltech prophet’s last generation of Examining the available numbers from several sources, researches in neuromorphic analog silicon now has Charlie Burger of the GTR affirms the Roberts estimates of enabled four major new ventures beyond Foveon, between 2.7 and 3.2 fold annual traffic growth (see chart including Synaptics, the global leader in touchpads, story). More importantly, he compares Roberts’ “Rule of Sonic Innovations (SNCI), the technology pioneer in Traffic Growth” doubling every seven months, with Moore’s

2 GILDER TECHNOLOGY REPORT law of electronic advance doubling every 18 months, and channel, was to use low-powered, broadband signals [Simon] Cao’s law of optical bandwidth doubling every five rather than “punching through” with high-powered, months. Measured by lambda bit miles (bitrate on a wave- narrowband signals. “Wide and weak” became the key length channel multiplied by the distance between opto- to the efficiency of CDMA, and to its present position electronic regenerators), optical capabilities are rising at of preeminence in wireless, with Sprint PCS (PCS) least at the Cao’s law pace. last month reporting 1.2 million new subscribers at ris- In 1995, a state-of-the-art Ciena (CIEN) WDM system ing average revenue per user (ARPU), and a 55 percent had four lambdas carrying 622 megabits per second 300 increase in year to year revenues. While rivals lust for miles, for a total of 746 Gbps-miles. Earlier this year more spectrum, Sprint’s migration path, the Broadwing deployed a Corvis (CORV) system capable of CDMA2000 3G system offers 153 Kbps of data and carrying 160 lambdas at 10 Gbps over 3,000 miles for a total conserves spectrum by a near doubling of voice capac- of 4.8 million Gbps-miles. This is 6,434 times growth in five ity. With pin compatible chips, forward and backward years, or a doubling every five months. The beat goes on. compatibility, and a software upgrade, Sprint’s new CIBC reports that Broadwing is now negotiating with generation network costs are trivial compared to its Corvis an increase in channel count from 160 to 320. rivals. Already in Korea the first CDMA2000 rollout Corvis reported at Telecosm that new and drastical- has garnered over a million users. ly cheaper Raman amplifiers would enable expansion of the fiber transmit window far beyond the current 50 Deflation destruction terahertz between 1,260 nanometers and 1,625 The technology paradigm is still ascendant and its nanometers possible with Lucent’s (LU) Allwave. The promise is greater than ever. During Telecosm and my new Raman window would open to 90 terahertz interviews and speeches following it, however, I came between 1,100 nanometers and 1,700 nanometers. to contemplate with new humility the ominous power Confirming this ever expanding horizon of optics was of Telecosm debt. In a high growth economy, leverage Terry Turpin of Essex (ESEX), who confirmed his claim was king so long as the federal reserve believed in in an earlier GTR (August 2001) that 16,000 lambdas growth and supplied the dollars to accommodate it. could be put stably and efficiently on a single fiber. After watching the triumphs of MCI, McCaw Cellular, Following Carver Mead’s suggestion at last year’s TCI, NewsCorp, Comcast (CCZ), Time Warner and Dynamic Silicon conference, Doug Lockie of Endwave other debt ridden suppliers of the infrastructure and (ENWV) contended at Telecosm that a better estimate of content of the 1990s information economy, I regarded Moore’s law would include the rapid acceleration of debt as a badge of entrepreneurial confidence, a prom- microprocessor clock rates. Up from 100 megahertz in ise of bold deployment, and a tool of tax avoidance. 1992 to 1.2 gigahertz this year and projected by Intel Rushing toward the Telecosm, impelled by Moore’s law (INTC) to a theoretical high of some 60 gigahertz over and Metcalfe’s law, a company without debt was like a the next decade, clock cycles are undeniably climbing Formula One racer confined to low gear. A company fast. But because the principal driver of clock speeds is without leverage was a triple taxed chump, maximizing the very density of circuits that Moore’s law describes, for Uncle Sam. I saw the debt defying audacity of simply adding acceleration to density entails considerable Bernie Ebbers of WorldCom, Bernie Schwartz of double counting. While both Roberts and Lockie urge us Globalstar/Loral, Steve Garofalo of Metromedia Fiber never to bet against Moore’s law, even a super Moore of and Gary Winnick of Global Crossing as a warrant of near annual doubling cannot keep up with Internet traf- their assurance and a tool of their sure success. I put fic expansion. Traffic will ultimately outpace any core most of my available family funds into Global Crossing. network technology dependent on electronics. I celebrated Greg Maffei of 360networks (TSIXQ). Although optics advances some 33 percent faster than When Joe Nacchio bought US West, I bumped him traffic does, Cao’s law of a five month optical doubling from the list. I did not foresee deflation. Since starv- rate does not apply to access technologies. On visits to ing an economy of liquidity and dooming most of its Agilent (A), Scale Eight, BlueArc, and Qualcomm entrepreneurial debtors is the most perverse of mone- (QCOM) I contemplated the residual scarcities of the tary policies, I could not imagine it in America. Telecosm. They sum up to lightspeed, power, and sili- Brother, was I ever wrong. con area, all converging tightly to constrain the domains of wireless bandwidth. Damodaran’s dark side I have sometimes implied a conflict between the PCS sprints ahead usual techniques of financial analysis and the paradigms Speaking at the CDMA conference in San Diego in of this letter, and not without reason. Most such tech- late October, I recalled the analysis of Andrew niques are either true but trivial, or nonsense. Most fun- Viterbi—creator of Viterbi codes and co-founder of damental analysis, for instance, is not fundamenatal at Qualcomm. Viterbi, following Shannon, argued the all and depends entirely on the same faith in the conti- best way to achieve a higher signal to noise ratio and nuity of trends that supports the chartist trumpery. In thus the highest effective transmission rate for any essence, prior earnings predict future earnings. True,

NOVEMBER 2001, VOLUME VI NUMBER 11 3 WASTELAND

asting portentous light on the future of the industry was the Internet Internet traffic still booms, traffic data of the top 19 carriers presented at Telecosm by Caspian’s Chart 1 CLarry Roberts. Chart 1 plots Roberts’ finalized traffic figures for April 100 2000 through April 2001. The figures used for April 2001 through October 2001 are based on Roberts’ preliminary findings for the last 6 months. Despite a number of short-lived periods of 4X, 3X, and 0X growth during the last 18 months, the slope of the trend line in chart 1 depicts a steady 3.2X overall annu- al growth rate, resulting in an estimate of 91 petabytes (PB) of Internet traffic in the month of October. In his presentation at Telecosm, Roberts quoted the same 3.2X annual growth rate figure, with the assertion that it has significant long-term validity. Is the 3.2X annual growth rate figure a valid long-term metric or is it simply an anomaly? Chart 2 extends our 3.2X trend line back to December ‘94, the last (PB/month) Traffic Internet month in which the National Science Foundation (NSF) actually measured and released Internet traffic figures. In doing so, we arrive at 33 terabytes per month of traffic, remarkably close to NSF’s measured figure of 16 ter- abytes per month, in December ‘94. We’re only three months off, in fact, 10 crossing the 16 TB/month point in September of ‘94 rather than December. Apr 00 Oct 00 Apr 01 Oct 01 Internet backbone traffic during the seven-year period from 1995 to pres- ent seems to be grown at a remarkably consistent rate, approximately dou- bling every seven months. But the NSF annual traffic growth rate of data from 1990 to 1994, yields a consistent 2X annual rate. What happened in 1995 to cause an but is the law: 3.2X yearly? abrupt growth rate increase? Browsers made the Net user friendly and Chart 2 email, AOL, and Compuserve moved onto the Net—an instant usage 100 explosion. The Internet established a new growth metric. Annual traffic surged to an unprecedented 9.7X yearly growth rate, doubling every 3.5 months, during the two-year period spanning 1995 and 1996. To further validate Roberts’ figures we looked to an ongoing Internet traf- 10 fic study conducted by University of Minnesota’s highly revered Internet traf- fic expert Andrew Odlyzko, former member of AT&T Labs’ research team, who estimated December ’00 backbone traffic to be somewhere between 20 and 35 PB per month. Roberts’ value is 31PB. Insomuch as Odlyzko is a firm believ- 1 er in a 2X annual long-term Internet traffic growth rate, one should note his remarkable December 1996 traffic figure, plotted in chart 3, along with his traffic-range estimates for subsequent Decembers from 1997 to 2000 (plotted as tick marks). Not only do Odlyzko’s traffic estimates overlap Roberts’ data for December 0.1 ’00, but they also overlap our original trend line for December ’99. Trendline extended back to December 1994

Figure approximated using Internet Traffic (PB/month) Furthermore, one can imagine a trend of alternating periods of rapid and Roberts’ data (0.033 PB/Month) sluggish growth (shown by the curved blue line) intersecting both Odlyzko’s NSF figure (0.016 PB/Month) and Roberts’ data and “settle” to reveal a constant 3.2X annual growth rate. 0.01 An infinite number of similar curves, however, could also fit the two data sets. Dec 94 Dec 96 Dec 98 Dec 00 Oct 01 Chart 4 shows a second example of such a curve. Considering the roughness of the “data” such a fine-tuned analysis is unjustifiable. Roberts extrapolated his figures based on carrier 95 percentile values, the accuracy of which depends on frequency of measurement and trusted sources hidden deep inside the highly furtive carriers. Odlyzko’s figures were derived from a variety of publicly available information sources, but only a handful of ISPs. A high level of uncertainty is indicated by his broad possible value ranges. Chart 5 shows a more reasonable linear trend, encompassing a period beginning in December ’96, when the Internet began to “settle down” from its growth spurt. The trend line in chart 5 fits both Roberts’ and Odlyzko’s data sets, revealing—again— a constant long-term growth trend, this one a 2.7X annual growth rate and thus a doubling of Internet traffic every 8.5 months. Whether you accept the a rate of traffic doubling every 7 months, or a more conservative 8.5 months, it is clear that Internet traffic is growing at a fairly constant exponential rate.

4 GILDER TECHNOLOGY REPORT Yes, says AT&T study. Or could it be 2X yearly? Chart 3 Chart 4

100 100

10

Odlyzko’s figure 1 10 Another example of a Apr 00 Oct 00 Apr 01 Oct 01 curve fitting the same two data sets

0.1 Internet Traffic Internet Traffic (PB/month) (PB/month) Internet Traffic

0.01 1 Dec 94 Dec 96 Dec 98 Dec 00 Oct 01 Dec 96 Dec 97 Dec 98 Dec 99

Burger confirms 2.7X, as industry “wastes” bandwidth. Chart 5 Chart 6 45% 100

ISP Average (Average of ISP Utilizations) 40%

35% 10 Linear trend line fitting Apr 00 Oct 00 Apr 01 Oct 01 the same two data sets over the same period.

Average of Totals 30% (Total Usage/Total Capacity)

2.7 fold annual traffic growth rate (PB/month) Internet Traffic

1 25% Dec 96 Dec 97 Dec 98 Dec 99 Jan 00 Apr 00 Jul 00 Oct 00 Jan 01 Apr 01 Jul 01 Oct 01

Sources: Roberts et al., Caspian Networks; K.G. Coffman and A.M. Odlyzko, AT&T Research Labs

Yet, Internet traffic is far from inelastic. Just as Moore’s law predicts a doubling of computer power every 18 months and so Cao’s law predicts the doubling of “optical power” every 5 months, so Roberts’s rule predicts a doubling of Internet traffic every 7 to 8.5 months. The price elasticity of demand for bandwidth is indicated not only by the amount of traffic but also by the total band- width capacity used to accommodate traffic. As computer chips grew more powerful our use of the chips did not keep up with the increase in power, we just wasted more and more transistors. The same trend is true for bandwidth. Average percentage bandwidth utilization rates plummet as bandwidth gets cheaper and we waste more and more of it (chart 6). This is the heart of the paradigm. - Charles Burger

NOVEMBER 2001, VOLUME VI NUMBER 11 5 but trivial, for very short time periods or for franchise investment, financial analysis provides no alternative to companies in industries of glacial stability. In other fundamental appraisal of technology paradigms. words, companies of little or no interest. Lengthen the The last year has taught us, however, that even time period or focus on more innovative industries and Telecosm vision needs to be quantitatively structured for the proposition that the past predicts the future particular companies. Damodaran’s model provides a struc- becomes ever more nonsensical. Complicate the theo- ture for appraisal of company values based on growth ry, add jargon, and smile knowingly, and it becomes expectations. Almost entirely subjective, expectations are nonsense on exponential stilts. All these models try to little correlated with historic trends. do without thinking. Rather than imagining the future they try to automate its prediction. Risky business Nevertheless, Telecosm analysis clearly would benefit The dominant fundamental or intrinsic models all from a model that could test the imagined future against hinge on the discounted present value of future cash whatever is quantifiable in the present. My favorite can- flows, which of course diminish sharply with risk. Yet didate for the modeling task comes from Aswath all investment theories concur that entrepreneurial Damodaran, whose The Dark Side of Valuation proposes wealth comes from acceptance of risks. Resolving the a valuation method almost entirely congruent and com- paradox, Damodaran defines risk as variability of plementary to the technology paradigm. returns or beta and cites evidence that high beta stocks Damodaran cites evidence (p149) of near zero cor- offer better returns than low beta stocks. Beta, howev- relation between historic and future growth rates of er, is variability in relation to some market index. It is earnings for most companies and even less correlation statistically definable and to some extent insurable for tech firms (p150), though revenues align somewhat through diversification. Entrepreneurial risk, as defined better with past trends. In other words a company that by Frank Knight in his canonical Risk, Uncertainty and is growing fast now is scarcely more likely to grow fast Profit, is uninsurable, because every instance is next year than a slow grower is. He shows that ana- unprecedented by definition. Entrepreneurship may be lysts, if anything, do worse than historic trendline defined as the launching of innovations, which are extension in projecting growth rates (pp156-158), and inherently unique, at least in some crucial property. that well regarded management is a bad investment Indeed, the more unique an innovation the more likely compared to badly regarded management (high regard it can confer a proprietary advantage, with barriers to being a reflection of the analyst’s predictions rather entry, that imparts outsized profits. But a unique than a determinant of them). And he shows that the instance creates uninsurable risk. This form of risk is more refinement of detail that enters a growth projec- absolute and irremediable. Thus, it should lift to near tion the less accurate it is likely to be. A general sense prohibitive levels the “risk premiums” employed in net of paradigms, as I might put it, beats an intricate analy- present value calculations of the worth of its specula- sis of every product and its prospects. tive cash flows. Such a method would block most inno- vation and thus most business profits and consequently Aswath Damodaran proposes a most stock market capital gains. It would orient the valuation method almost entirely investor precisely toward those projects least likely to yield a profit beyond the risk free interest rate. congruent and complementary to Damodaran’s solution is a “real option” model. The the technology paradigm investment is gauged neither by the calculable risk adjusted valuation of its projected cash flows, nor by its Then, in a dramatic coup, Damodaran proves his opportunity costs as measured by the rewards of other rules by offering his own (rather detailed) projections safer ventures, but by the opportunity payoff: the for Cisco (CSCO), Amazon (AMZN), Motorola options opened up to pursue unique goods and servic- (MOT), Ariba (ARBA), and Rediff (REDF). Issued in es or to pioneer in contiguous markets or to engineer June 2000, Damodaran’s projections correctly predicted related innovations. Amazon’s Latin American venture, a slump of valuations from their year 2000 peaks, but for example may well destroy value in discounted cash failed utterly to anticipate the cliffs of 2001. flow terms while at the same time buying an “option” Damodaran did not correct for deflation either. He to enter the Latin American market in a broader range assigned a value of $44.92 to Cisco, now at $16.08; of businesses, when the opportunity presents itself. $32.39 to Motorola, now at $17.19; $34.37 to Amazon, The “real option” equation is formidably complex, now at $7.87; $72.13 to Ariba, now at $2.31; and involving the creation of simulated or replicated set of $19.05 to Rediff, now at 75 cents. I do not list these risk and reward components, estimation of the current numbers to gloat. In my fiber glass house, I am in no value of the underlying asset, and the variation in that position for Shadenfreude. Already amply stoned and value. The central insight is that for technology compa- deflated, I have no grounds from which to hurl any nies risk is defined as variability of returns—the dynamic smithereens. But whether discounted cash, relative range of outcomes—and this risk index will reduce the multiples, economic value added, or cash flow return on net present value of existing assets, but raise the value of

6 GILDER TECHNOLOGY REPORT real options. Representing opportunities and potential CEO Legere says it begins now. If it does, two fac- projects, they can be modeled the same way as finan- tors will control GX’s growth. The first is the severe cial options or “calls.” They combine the cost of the challenge of signing up as customers the Forbes 500 option, which is the expense of staying in the business global companies who are the world’s biggest users of or purchasing the license or patent or continuing the telecom and Internet services but have large, compli- R&D, with the strike price, which is the cost of actu- cated, facilities-based, IT-entrenched contracts with the ally entering production, launching the product or current telecom players. Not easy. commercializing a development or project. The net Facilitating the job of GX’s sales force is the threefold payoff is the yield of the asset minus the cost of the annual pace of Internet traffic growth, which will force option. Cisco, by this standard, is chiefly an aggrega- such customers to seek as a second source anyone with tor of options (commercializer of acquired projects) global capacity. Assuming a) that GX survives its cash and its long outsized returns benefited from risk. crunch long enough to exercise its option, b) that the ini- Thus, a discounted present value calculation, with a tial sales job is tough, and c) that traffic growth both trig- risk penalty, is an inappropriate gauge of Cisco’s worth. gers the initial sales and drives out- year revenue growth, As an aggregator of options Cisco can flexibly exploit a GX’s growth curve could vary dramatically from range of networking technologies Damodaran’s suggestion. that are not yielding significant Given all these factors, includ- revenue today. Some options will Global Crossing Growth Path ing a world economy that will take remain under water (Monterey Figure 1 some time regaining its health, we Networks). But most will pay off to Year Damodaran GTR offer this alternative (see figure some degree (Cerent, Arrowpoint, 2001 0% 0% 1). Plugging different growth Andy Bectolsheim’s Granite for 2002 40% 30% assumptions into Damodaran’s gigabit Ethernet). 2003 30% 40% model, Bret came up with a valu- 2004 20% 40% ation of $19.9 billion or a stock GX’s real business begins 2005 10% 30% price of $21.56. Damodaran provides a practical 2006 10% 30% guide for valuing several of the 2007 10% 20% Dynegy challenges GX Telecosm companies. He even sup- 2008 8% 10% Assuming it survives the next plies a specific example in his 2009 6% 10% 18 months, the chief challenge for appraisal of Motorola, which he 2010 5% 5% Global Crossing will be technology saw as undervalued. As we went to Compound Avg management. Advancing at an ever press, he also gave us an appraisal Growth Rate 10% 17% accelerating pace, the optical capa- of Global Crossing that valued the Share Price $3.22 $21.56 bilities of the Global Crossing net- company at $2.9 billion ($3.22 per Equity Value $2.9b $19.9b work will be available at far cheap- share) at a time it was valued by er prices to new entrants. One, the the market at $1.52 billion dollars well financed power company or roughly $1.67 a share. Bret Swanson challenged Dynegy (DYN) of Houston (now maneuvering to buy some of his premises. Key to Damodaran’s model are Enron) has contrived through a combination of leases revenue growth assumptions, and with the benefit of and buildouts a global network over the last year that in hindsight and a paradigm, we see reason to question some respects already rivals Global Crossing’s. Using his findings in several important respects. Tellium (TELM) optical switches and Fujitsu (FJTSY) Damodaran’s suggested growth path for GX starts at advanced WDM multiplexing–reportedly based on the zero percent in year one (2001) and progresses through Avanex (AVNX) PowerMux–Dynegy has 16,000 miles of 40, 30, 20, 10, 10, 10, 8, 6, and 5 for the next nine years. domestic routes and an emerging global presence. Through the sale of undersea IRUs (simple capacity Dynegy’s network could become fully competitive with mostly across the Atlantic), GX achieved a billion dollars Global Crossing’s if the company does not adopt new in sales faster than any other company in history. Then technologies aggressively. As the Roberts trend of annual increased competition slowed growth rates to a still-blis- tripling of Internet traffic extends around the globe, John tering 40 percent from more than 100 percent. But the Legere must make his company a worldwide leader in “real option” created by GX’s so far uniquely seamless testing and applying the new lambda technologies and global network is not Atlantic IRU sales, which were a integrating them with its current industry leading stop-gap measure to finance the network. The “real IP/MPLS (multiprotocol label switched) offerings. option,” the real business, is the ability to offer IP net- work services to global enterprises without the burden of management and transaction costs entailed in network border crossings. It was precisely to create this option that the network was built. Until it has been “exercised” - George Gilder GX’s business can hardly be said to have begun. November 20, 2001

NOVEMBER 2001, VOLUME VI NUMBER 11 7 TELECOSM TECHNOLOGIES ASCENDANT TECHNOLOGY COMPANY (SYMBOL) REFERENCE OCT ‘01: 52 WEEK MARKET DATE / PRICE MONTH END RANGE CAP FIBER OPTICS Optical Fiber, Photonic Components Corning (GLW) 5/1/98 13.64 8.06 6.92 - 79.75 7.6B Wave Division Multiplexing (WDM) Components JDS Uniphase (JDSU) 6/27/97 3.63 8.09 5.12 - 83.00 10.7B Adaptive Photonic Processors Avanex (AVNX) 3/31/00 151.75 4.96 2.70 - 108.63 329.5M All-Optical Cross-Connects, Test Equipment Agilent (A) 4/28/00 88.63 22.27 18.00 - 68.00 10.26B Tunable Sources and WDM Components New Focus (NUFO) 11/30/00 20.31 2.90 2.10 - 68.00 218.9M Crystal-Based WDM and Optical Switching Chorum (private) 12/29/00 — — — — WDM Metro Systems ONI (ONIS) 12/29/00 39.56 4.91 3.50 - 82.50 677.3M WDM Systems, Raman Corvis (CORV) 3/30/01 7.03 2.24 1.19 - 68.25 806.6M Metro Semiconductor Optical Amplifiers Genoa (private) 3/30/01 — — — — Optical Processors Essex (ESEX.OB) 7/31/01 5.90 5.85 1.50 - 6.70 29.0M

LAST MILE Cable Modem Chipsets, Broadband ICs Broadcom (BRCM) 4/17/98 6.00* 34.41 18.40 - 227.50 8.9B S-CDMA Cable Modems Terayon (TERN) 12/3/98 15.81 11.35 2.36 - 25.81 776.2M Linear Power Amplifiers, Broadband Modems Conexant (CNXT) 3/31/99 13.84 10.15 6.57 - 33.50 2.6B Broadband Wireless Access, Network Software Soma Networks (private) 2/28/01 — — — — WIRELESS Satellite Technology Loral (LOR) 7/30/99 18.88 1.27 1.03 - 6.56 423.3M Low Earth Orbit Satellite (LEOS) Wireless Transmission Globalstar (GSTRF) 8/29/96 11.88 0.53 0.20 - 3.94 58.5M Code Division Multiple Access (CDMA) Chips, Phones Qualcomm (QCOM) 7/19/96 4.75 49.05 38.31 - 107.8 37.3B Nationwide CDMA Wireless Network Sprint (PCS) 12/3/98 7. 1 9 * 22.30 15.72 - 38.44 20.9B CDMA Handsets and Broadband Innovation Motorola (MOT) 2/29/00 56.83 16.37 10.50 - 26.25 36.1B Wireless System Construction and Management Wireless Facilities (WFII) 7/31/00 63.63 6.59 3.31 - 54.75 297.9M GLOBAL NETWORK Metropolitan Fiber Optic Networks Metromedia (MFNX) 9/30/99 12.25 0.73 0.25 - 23.50 447.5M Global Submarine Fiber Optic Network Global Crossing (GX) 10/30/98 14.81 1.13 0.38 - 25.88 1.0B Regional Broadband Fiber Optic Network NEON (NOPT) 6/30/99 15.06 3.07 2.16 - 19.94 65.5M National Lambda Circuit Sales Broadwing (BRW) 6/29/01 24.45 9.26 8.82 - 28.88 2.0B Internet Backbone and Broadband Wireless Access WorldCom (WCOM) 8/29/97 19.95 13.46 11.50 - 25.69 39.8B STOREWIDTH Java Programming Language, Internet Servers Sun Microsystems(SUNW) 8/13/96 6.88 10.15 7.52 - 57.49 32.9B Network Storage and Caching Solutions Mirror Image (XLA) 1/31/00 29 .00 1.84 1.00 - 15.25 208.8M Remote Storewidth Services StorageNetworks (STOR) 5/31/00 27.00* 5.06 3.65 - 73.69 490.3M Hardware-centric Networked Storage BlueArc (private) 1/31/01 — — — — Virtual Private Networks, Encrypted Internet File Sharing Mangosoft (MNGX.OB) 1/31/01 1.0 0 0.84 0.34 - 5.69 22.7M Massively Parallel Global Storewidth Solutions Scale Eight (private) 8/31/01 — — — — MICROCOSM Analog, Digital, and Mixed Signal Processors Analog Devices (ADI) 7/31/97 11. 19 37.98 29.00 - 68.00 13.7B Silicon Germanium (SiGe) Based Photonic Devices Applied Micro Circuits (AMCC) 7/31/98 5.67 11.04 6.01 - 88.25 3.3B Programming Logic, SiGe, Single-Chip Systems Atmel (ATML) 4/3/98 4.42 7.96 5.48 - 18.44 3.7B Single-Chip ASIC Systems, CDMA Chip Sets LSI Logic (LSI) 7/31/97 15.75 16.95 9.78 - 33.00 6.2B Single-Chip Systems, Silicon Germanium (SiGe) Chips National Semiconductor (NSM) 7/31/97 31.50 25.98 17.13 - 35.10 4.6B Analog, Digital, and Mixed Signal Processors, Micromirrors Texas Instruments (TXN) 11/7/96 5.94 27.99 20.10 - 54.69 48.5B Field Programmable Gate Arrays (FPGAs) Xilinx (XLNX) 10/25/96 8.22 30.50 19.52 - 74.56 10.2B Seven Layer Network Processors EZchip (LNOP) 8/31/00 16.75 5.40 2.70 - 31.25 34.8M Network Chips and Lightwave MEMS Cypress Semiconductor (CY) 9/29/00 41.56 19.75 13.72 - 39.69 2.4B Field Programmable Gate Arrays (FPGAs) Altera (ALTR) 1/31/01 30.25 20.20 14.66 - 42.00 7.8B * INITIAL PUBLIC OFFERING NOTE: The Telecosm Table is not a model Gilder Technology Report Published by Gilder Publishing, LLC and Forbes Inc. portfolio. It is a list of technologies in the 291A Main Street • Great Barrington, MA 01230 FOR SUBSCRIPTION Gilder Paradigm and of companies that lead INFORMATION in their application. Companies appear on Tel: (888)484-2727 • Fax: (413)644-2123 • Email: [email protected] TELEPHONE this list only for their technology leadership, EDITOR: George Gilder without consideration of their current share TOLL FREE: price or the appropriate timing of an invest- PUBLISHER: Richard Vigilante ment decision. The presence of a company ANALYSTS: Charles Burger, Mary Collins, Bret Swanson on the list is not a recommendation to buy RESEARCH ASSISTANT: John Hammill (888) 484-2727 shares at the current price. Reference Price MANAGING EDITOR: Debi Kennedy is the company’s closing share price on the Website: DESIGNER: Julie Ward Reference Date, the day the company was www.gildertech.com added to the table, typically the last trading SUBSCRIPTION DIRECTOR: Rosaline Fernandes day of the month prior to publication. Mr. Copyright © 2001, by Gilder and other GTR staff may hold posi- GENERAL MANAGER, GILDER TECH.COM: David S. Dortman Gilder Publishing, LLC tions in some or all of the stocks listed. PRESIDENT: Mark T. Ziebarth